Title: “A Common Perception and Footstep Planning Approach for Different Humanoid Robots with Application to the DRC.”
Abstract:A Common Perception and Footstep Planning Approach for Different Humanoid Robots with Application to DRC Scenarios
In  we have presented a footstep planning approach originally developed for Atlas from Boston Dynamics Incorporated (BDI). This approach is able to utilize the versatile capabilities of humanoid walking in challenging terrain. Since 2012 various new full-size humanoid robots, like the 2nd version of Atlas, THOR-MANG (ROBOTIS) or ESCHER (VirginiaTech) have been introduced to participate in the DARPA Robotics Challenge (DRC) Finals in 2015. All of them are delivered with their own controllers for basic walking motions, but without any footstep planning. This raised the opportunity to demonstrate that our approach developed in  can be applied to other full-size humanoid robots as it can cope with the robot specific walking controller provided as basically black boxes. The further developed footstep planning framework has been successfully applied to Atlas, THOR-Mang and ESCHER humanoid robots in real world applications. The recent framework, which will be presented in the talk, is now ready to use a very broad class of humanoid robots as long as a basic controller for walking and stepping motions is provided.
Our main objective is to provide a versatile and highly capable, but also easy integrable as well as expandable footstep planning framework in ROS using search-based algorithms (like A*). Any user of the framework should only have to implement and extend robot specific elements to get the advanced planning system running instead of developing a modified version of an existing planning system or even starting from scratch each time. All already implemented, tested and proven algorithms are kept untouched which decreases the possibilities of errors and saves a lot of implementation effort. Although the framework must generalize well, it is able to solve difficult terrain task problems and utilize the versatile locomotion capabilities of the given walking controller. Hereby, the framework comes with an online 3D terrain generator which has also been applied and validated successfully for real world scenarios.
The flexibility of the footstep planning framework is provided by plugin as well as parameter management system. Therefore, the footstep planning processing chain has been analyzed on common places where a user might want to modify the behavior of the planner (e.g. collision checking). For each identified place a plugin has been introduced which may be (un)loaded during runtime from/into the planner. These plugins are supposed to inject efficiently user specific code into the planning system without any modification to the framework itself. As every user made code needs its own parameters to run properly, a parameter management system has been introduced as well. This system is able to overcome the basic conflict of rigid message types needed by ROS for interprocess communication and the need of flexible content of parameter sets due to user defined parameters.
The integrated footstep planning framework is designed to be deployed quickly into an existing ROS setup. For this purpose the framework comes already with many basic as well as advanced plugins which enable a quick start for 3D footstep planning with a new humanoid robot. Also through a number of rqt widgets a basic graphical user interface is provided.
Up to now this approach has been integrated successfully for in total five unique walking controllers of the three different full-size humanoid robots Atlas, THOR-Mang and ESCHER. It has enabled them to tackle challenging terrain scenarios as occurred in the DRC. In our talk we will explain the basics of the novel footstep planning framework and show an advanced implementation example with Atlas which is capable to perform full 3D footstep planning as demonstrated in the videos  and .
 A. Stumpf, S. Kohlbrecher, D.C. Conner, O. von Stryk; Supervised Footstep Planning for Humanoid Robots in Rough Terrain Tasks using a Black Box Walking Controller; In Proc. IEEE-RAS Intl. Conf. Humanoid Robots, pp. 287-294, Nov 18-10, 2014